fixed incorrect implementation of bag distance (reported by r. feldt)

src/measures/dist_bag.c

@@ -103,7 +103,7 @@ static void bag_destroy(bag_t * xh)
  */
 float dist_bag_compare(hstring_t x, hstring_t y)
 {
-    float d = 0;
+    float xd = 0, yd = 0;
     bag_t *xh, *yh, *xb, *yb;
 
     xh = bag_create(x);
@@ -113,18 +113,20 @@ float dist_bag_compare(hstring_t x, hstring_t y)
     for (xb = xh; xb != NULL; xb = xb->hh.next) {
         HASH_FIND(hh, yh, &(xb->sym), sizeof(sym_t), yb);
         if (!yb) {
-            d += xb->cnt;
+            xd += xb->cnt;
         } else {
-            d += fabs(xb->cnt - yb->cnt);
+            float diff = xb->cnt - yb->cnt;
+            xd += fmax(+diff, 0);
+            yd += fmax(-diff, 0);
             missing -= yb->cnt;
         }
     }
-    d += missing;
+    yd += missing;
 
     bag_destroy(xh);
     bag_destroy(yh);
 
-    return lnorm(n, d, x, y);
+    return lnorm(n, fmax(xd, yd), x, y);
 }
 
 /** @} */

tests/check_measures.txt

@@ -1,12 +1,12 @@
 dist_bag
-0,12,19,15,27,30,18,15
-12,0,19,15,23,30,18,19
-19,19,0,18,26,23,23,18
-15,15,18,0,22,31,19,16
-27,23,26,22,0,37,27,28
-30,30,23,31,37,0,32,29
-18,18,23,19,27,32,0,21
-15,19,18,16,28,29,21,0
+0,9,11,12,16,24,9,10
+9,0,14,9,17,27,12,15
+11,14,0,15,14,19,13,10
+12,9,15,0,18,29,14,15
+16,17,14,18,0,25,16,14
+24,27,19,29,25,0,25,21
+9,12,13,14,16,25,0,13
+10,15,10,15,14,21,13,0
 dist_compression
 0.12,0.68,0.5893,0.68,0.7143,0.8049,0.68,0.6833
 0.68,0.1053,0.7143,0.5789,0.7143,0.8049,0.68,0.7333
@@ -89,14 +89,14 @@ dist_osa
 16,14,17,16,19,29,0,19
 18,16,17,15,21,24,19,0
 kern_distance
-289,133,164,64,22,307,127,274
-133,121,80,-20,38,223,43,122
-164,80,400,70,104,548,80,280
-64,-20,70,64,32,164,-4,146
-22,38,104,32,484,170,22,92
-307,223,548,164,170,1225,245,434
-127,43,80,-4,22,245,289,166
-274,122,280,146,92,434,166,484
+289,164.5,284,104.5,258.5,469,248.5,336.5
+164.5,121,162.5,52,158,308.5,133,190
+284,162.5,400,119.5,344,632,260,392
+104.5,52,119.5,64,112,224,78.5,161.5
+258.5,158,344,112,484,542,258.5,386
+469,308.5,632,224,542,1225,444.5,634
+248.5,133,260,78.5,258.5,444.5,289,302
+336.5,190,392,161.5,386,634,302,484
 kern_spectrum
 15,0,3,0,0,0,0,1
 0,9,0,0,0,0,0,0

tests/dist_bag.c

@@ -39,26 +39,31 @@ struct hstring_test tests[] = {
     {"a", "a", "", 0},
     {"ab", "ba", "", 0},
     {"bab", "ba", "", 1},
-    {"abba", "babb", "", 2},
-    {"a.b", "a.c", "", 2},
-    {".a.b.", "a..c.", "", 2},
+    {"abba", "babb", "", 1},
+    {"a.b", "a.c", "", 1},
+    {".a.b.", "a..c.", "", 1},
     /* Comparison using tokens */
     {"", "", ".", 0},
     {"a", "", ".", 1},
     {"", "a", ".", 1},
     {"a", "a", ".", 0},
-    {"ab", "ba", ".", 2},
-    {"bab", "ba", ".", 2},
-    {"abba", "babb", ".", 2},
-    {"a.b", "a.c", ".", 2},
-    {".a.b.", "a..c.", ".", 2},
+    {"ab", "ba", ".", 1},
+    {"bab", "ba", ".", 1},
+    {"abba", "babb", ".", 1},
+    {"a.b", "a.c", ".", 1},
+    {".a.b.", "a..c.", ".", 1},
     /* Further test cases */
-    {"abcd", "axcy", "", 4},
-    {"abc", "axcy", "", 3},
-    {"abcd", "xcy", "", 5},
+    {"abcd", "axcy", "", 2},
+    {"abc", "axcy", "", 2},
+    {"abcd", "xcy", "", 3},
     {".x.y.", ".x.y.", ".", 0},
     {"x...y..", "...x..y", ".", 0},
     {".x.y", "x.y.", ".", 0},
+    /* Examples from paper by Bartolini et al. */
+    {"spire", "fare", "", 3},
+    {"fare", "spire", "", 3},
+    {"spire", "paris", "", 1},
+    {"paris", "spire", "", 1},
     {NULL}
 };